Pika vs Runway for characters: four engines in one credit currency, priced per second

The Pika-versus-Runway debate online is a two-engine debate held in two different wallets. Pika's credit, Runway's credit, each pricing a second against its own meter, each hiding the comparison the question is actually asking: for a one-character shot where identity has to hold across cuts, which second is the cheapest finished second?

The ranking pages (aitoolsbakery, aicomparison, saascrmreview, pickaistool, fluxnote, zsky, neuronad, vijaytalksai — all crawled 2026-10-20) answer "Runway for character, Pika for cost" in some variation. That is useful if you were only going to pick one of those two. It is useless if your shelf also carries Kling and Seedance, where the character answer is different again and the cost answer is different again.

I run a studio where the four engines sit side by side on one credit meter. Here is the four-way table nobody else publishes, because nobody else has the four engines on one wallet.

The engines, side by side, as of 2026-10-20

Pika and Runway rates below are off public SERP writing (sources dated 2026-10-20 in reports/seo/research.md); Kling O3 Pro and Seedance 2.5 are our shelf's own published rates off lib/models.ts.

engine	720p rate (public)	720p rate (our credits)	1080p rate (our credits)	character consistency	shot cap
Pika 2.5 (image→video)	$0.057/s @ 720p (as of 2026-10-20)	— (not on our shelf)	—	"improving but still fragile" per 2026-10 SERP writing	5s/10s single-shot
Runway Gen-4.5	$0.288/s @ 720p (Standard tier, as of 2026-10-20)	— (not on our shelf)	—	reference-driven, holds across shots	5s or 10s
Kling O3 Pro	—	24 silent / 30 voiced	— (not served)	elements lock (@Element1 … @Element5), the strongest cross-shot ID lock on our shelf	15s per generation
Seedance 2.5 (via Segmind)	—	70 (voiced=silent)	130	up to 30 image refs, by position	one pass, no join
Veo 3.1 Fast	—	20 silent / 30 voiced	20 silent / 30 voiced	three ingredient images, 8s only	8s (enum: 4, 6, 8)

A credit on our shelf is 1.0¢ on Lite ($10 / 1,000 credits), 0.75¢ on Creator ($45 / 6,000), 0.55¢ on Studio annual (effective). That gives the fourth column of the table most writers skip:

engine	credits/s voiced 720p	¢ on Lite	¢ on Creator	¢ on Studio annual
Kling O3 Pro	30 cr/s	30.0 ¢/s	22.5 ¢/s	16.5 ¢/s
Veo 3.1 Fast	30 cr/s	30.0 ¢/s	22.5 ¢/s	16.5 ¢/s
Seedance 2.5 720p	70 cr/s	70.0 ¢/s	52.5 ¢/s	38.5 ¢/s
Pika 2.5 720p	—	—	—	— ($0.057/s API-side)
Runway Gen-4.5 Standard	—	—	—	— (~$0.288/s off public writing)

Pika is the cheapest second in the room, when the brief is one shot where the character does not have to hold across cuts. Runway is the character answer the SERP picks for cross-shot ID. On our shelf the answer is different: Kling O3 Pro voiced 720p is 22.5¢/s on Creator ($0.225/s), which is cheaper per voiced second than Runway Gen-4.5 Standard's $0.288/s and more character-stable than Pika at $0.057/s. Pika stays cheapest where identity does not matter; Kling takes the second spot of "cheap AND character-stable" because the shelf prices against ByteDance and Kuaishou rather than Pika's own compute.

Character consistency, in the four frames

Pika 2.5. The ranking pages concede this is still the fragile one. The 2026-10 writing on Pika says "improving" without quoting a number. For a short single-shot where character continuity does not have to survive a cut, Pika is a line in the budget. For a six-shot story, every writer flags that the face drifts.

Runway Gen-4.5. The reference-driven workflow is good. Multiple SERP pages (aicomparison 2026-10-20) credit Runway with holding identity across shots for narrative video. The honest footnote is that Runway's "character" is a reference the engine can see in-prompt, close in function to what Kling's @Element tokens do without naming them.

Kling O3 Pro. Elements, up to five, each addressed by name in the prompt: @Element1 enters frame, @Element2 waits by the window. Two camera moves, two cuts, same identity because the engine is holding five specific elements against the specific tokens the prompt names. The one engine on our shelf whose references are addressed by name.

Seedance 2.5. Thirty image references, addressed by position (@Image1, @Image2 ...). Different workflow — no in-prompt naming, you place the images in order and refer to the position. Where the scene has 6+ characters this is the row the refusal copy sends to.

Veo 3.1 Fast. Three "ingredient" images, referenced in the prompt as plain words ("image 1", "image 2", "image 3") because Google has not shipped an addressing token. Reference-mode length is pinned to 8 seconds; the picker enforces it because the API will render 8 and bill 8 whatever the slider shows.

For a cross-shot same-character story, the shelf's rank order is: Kling O3 Pro (named elements, 15s per generation, 30 voiced / 24 silent) → Seedance 2.5 (30 references, one unbroken pass, 70 voiced=silent) → Veo 3.1 Fast (3 references, 8s pinned, 30 voiced / 20 silent) → Pika (one shot at a time, cheapest second). Runway Gen-4.5 sits in the middle on consistency and at the top on dollar cost; our Kling row is the alternative that keeps the identity and beats Runway on $/s on Creator.

The 20-second same-character scene, priced four ways

Four shots of the same character — opener, mid-shot, close-up reaction, close-out — 20 seconds total, voiced, 720p.

version	engine	credits	$ on Lite	$ on Creator	$ on Studio annual
Kling-only, our shelf	Kling O3 Pro voiced 720p, 4 shots × 5s	600	$6.00	$4.50	$3.30
Hybrid, our shelf	1 Veo 6s voiced + 3 Kling 4-5s voiced	180 + 420 = 600	$6.00	$4.50	$3.30
Seedance 2.5 one-take	20s voiced 720p	1,400	$14.00	$10.50	$7.70
Runway Gen-4.5 (public)	20s × 12 cr/s on Runway's meter	240 Runway credits ≈ $5.76 (Standard $0.288/s)	—	—	—
Pika 2.5 (public)	20s × 4 cr/s @ 720p on Pika's meter	80 Pika credits ≈ $1.14 (if $0.057/s holds)	—	—	—

Pika at $1.14 for 20 seconds wins on dollars, if the shots do not have to be the same character. For the same-character brief, Pika's "improving" rating in the SERP writing means the comparison that matters is Kling vs Runway, and the Kling number on Creator ($4.50) beats Runway Gen-4.5 Standard ($5.76). On Studio annual the gap widens: $3.30 vs $5.76, roughly 42% less per voiced same-character second.

That is the number no cross-vendor comparison page ships, because no competitor has Kling, Seedance, Veo and the Lite/Creator/Studio currencies on one rate card.

The decision we actually make per shot

One short shot, character does not need to carry across cuts, cost rules: Pika on its own meter (not on our shelf; included for budget framing). Our Veo 3.1 Fast silent 720p at 20 cr/s is the inside-the-shelf answer.

Multi-shot story where identity carries, up to 5 characters, under 15s per generation: Kling O3 Pro.

Multi-shot story with 6+ characters, or a 15–30s one-take without a join: Seedance 2.5.

8-second native-audio opener with clean place and light compute: Veo 3.1 Fast.

Draft pass of a Seedance 2.0 scene before the final: Seedance 2.0 Fast at half the price.

The two-engine debate the SERP has was decided inside the two engines' wallets. The four-engine answer is a different arithmetic: Kling where identity has to hold, Veo where native audio has to land in-pass, Seedance 2.5 where the shot is wide or long, Pika and Runway not on our shelf because the two rows the shelf carries beat them on either consistency or cost. Our 300-credit free account is enough to burn 10 Veo seconds plus 5 Kling seconds — the smallest two-engine test you can run against your own character. Try the four-way quote on a scene you already know.

The thing worth comparing is not Pika vs Runway. It is the four engines priced in one credit currency, which is a thing only a shelf like ours can publish.